Neural network–enabled accelerated discovery of multifunctional metamaterials for adaptive multispectral stealth applications

Abstract

The development of advanced multispectral compatible stealth materials (CSMs) based on metamaterials faces significant challenges, including computational inefficiency, prohibitive costs, and the persistent issue of local optima in conventional design approaches. This study presents a transformative inverse design framework that revolutionizes the field by enabling rapid optimization within a quasi-infinite solution space. Departing from traditional low-dimensional design paradigms that are constrained by limited solution spaces and excessive reliance on manual intervention, our innovative approach introduces three key advancements: (1) a randomized cut-line coding methodology that generates an expansive, high-dimensional design space capable of addressing diverse stealth requirements; (2) a novel hybrid intelligence system combining genetic algorithms with neural networks for unprecedented computational efficiency and design flexibility; and (3) a multilayer architecture integrating conductive surface materials that achieves remarkable multispectral performance. The resulting CSMs, with a mere 1.24 mm thickness and 2.22 kg/m² surface density, demonstrate exceptional capabilities, including ultrabroadband antireflection (reflectivity <0.1 across 8.9–18 GHz), dynamic multiband performance modulation (tunable within 6–18 GHz), radar cross-section reduction, and beam deflection - all programmable through customized fitness functions. Furthermore, the materials exhibit superior infrared stealth characteristics, achieving emissivity values as low as 0.3. This work establishes a new paradigm for the development of adaptive multispectral stealth materials, offering unprecedented versatility in diverse detection environments.